$$\Omega_1=\left\{c\in\Bbb C:\begin{bmatrix}1&c\\\bar c&1\\ \end{bmatrix}\text{ is non-negative definite } \right\} $$

$$\Omega_2= \left\{c\in\Bbb C: \begin{bmatrix} 1 & c & c \\ \bar c & 1 & c \\ \bar c & \bar c & 1 \\ \end{bmatrix} \text{ is non-negative definite } \right\}$$

Let $$\bar D=\{z\in\Bbb C:|z|\le1\}$$ Then 1) $\Omega_1=\bar D,\Omega_2=\bar D$

2). $\Omega_1\neq\bar D, \Omega_2=\bar D$

3). $\Omega_1=\bar D, \Omega_2\neq\bar D$

4). $\Omega_1\neq\bar D, \Omega_2\neq\bar D$

How to proceed? Thank you.

  • $\begingroup$ First, recall the definition of a non-negative definite matrix. What conditions does this put on $c$ in each case? $\endgroup$ – Alex G. Jun 21 '15 at 17:48
  • 2
    $\begingroup$ If these were real numbers, the first thing I would think of is that if the variance of each of two or three random variables is $1$, then the covariance is the correlation, and the correlation is always in $[-1,1]$, so $-1\le c\le 1$. Here one relies on the fact that the set of all matrices of covariances is precisely the set of all non-negative-definite matrices. My first guess is $|c|\le 1$ for both $\Omega_1$ and $\Omega_2$. ${}\qquad{}$ $\endgroup$ – Michael Hardy Jun 21 '15 at 17:53
  • 1
    $\begingroup$ non negative means positive semi definite i.e. all eigen values are ≥0 and atleast one eigen value 0 $\endgroup$ – Chiranjeev_Kumar Jun 21 '15 at 17:54
  • $\begingroup$ Just as a matter of style, I'd rather write $\{c\in\mathbb C:|c|\le 1\}$, using $c$ rather than $z$ as in the definitions of the other two sets. ${}\qquad{}$ $\endgroup$ – Michael Hardy Jun 21 '15 at 17:57
  • 1
    $\begingroup$ @Michel Hardy : if we take $c=-1$ then 3by3 matrix is non singuler with determinant value -4 which contradicting the given condition $\endgroup$ – Chiranjeev_Kumar Jun 21 '15 at 18:24
  1. Note that the matrices are Hermitian, so it is enough to check if the eigenvalues $\lambda\geq 0$ are non-negative, or equivalently, $$\mu~:=~1-\lambda~\leq~ 1.$$
  2. The characteristic polynomials read $$ p_1(\lambda) ~=~\mu^2-|c|^2, $$ and $$ p_2(\lambda) ~=~\mu^3+|c|^2(2{\rm Re}(c) -3\mu), $$ respectively.

  3. Define polar decomposition $c~=~re^{i\theta}~\in~\mathbb{C}$.

  4. The roots are $$ \mu~=~\pm |c|,$$ and $$ \mu = 2 r \cos\frac{\theta+2\pi p}{3},\qquad p\in\mathbb{Z},$$ respectively.

  5. Hence, $$ \Omega_1~=~\{c\in \mathbb{C} \mid |c|\leq 1\}~=~\bar{D}, $$ while $$ \Omega_2~=~\{re^{i\theta}\in \mathbb{C} \mid \forall p\in \mathbb{Z}:~ 2 r \cos\frac{\theta+2\pi p}{3}\leq 1\}~\neq~\bar{D}. $$ It is straightforward to check that $$ \{c\in \mathbb{C} \mid |c|\leq \frac{1}{2}\}~\subsetneq~\Omega_2~\subsetneq~\bar{D}. $$

The $2\times 2$ matrix has eigenvalues $1 \pm |c|$, so it is positive semidefinite if and only if $|c|\leq 1$.

For the $3\times 3$ matrix, observe that the determinant is $-4$ when $c=-1$, so the matrix is not positive semidefinite in this case.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.